#include <unistd.h>
#include <fcntl.h>
#include <xf86drm.h>
-#include <drm_fourcc.h>
+#include "drm-uapi/drm_fourcc.h"
#include "anv_private.h"
#include "util/strtod.h"
va_end(args);
}
-static VkResult
-anv_compute_heap_size(int fd, uint64_t gtt_size, uint64_t *heap_size)
+static uint64_t
+anv_compute_heap_size(int fd, uint64_t gtt_size)
{
/* Query the total ram from the system */
struct sysinfo info;
*/
uint64_t available_gtt = gtt_size * 3 / 4;
- *heap_size = MIN2(available_ram, available_gtt);
-
- return VK_SUCCESS;
+ return MIN2(available_ram, available_gtt);
}
static VkResult
device->supports_48bit_addresses = (device->info.gen >= 8) &&
gtt_size > (4ULL << 30 /* GiB */);
- uint64_t heap_size = 0;
- VkResult result = anv_compute_heap_size(fd, gtt_size, &heap_size);
- if (result != VK_SUCCESS)
- return result;
+ uint64_t heap_size = anv_compute_heap_size(fd, gtt_size);
if (heap_size > (2ull << 30) && !device->supports_48bit_addresses) {
/* When running with an overridden PCI ID, we may get a GTT size from
device->has_context_isolation =
anv_gem_get_param(fd, I915_PARAM_HAS_CONTEXT_ISOLATION);
- bool swizzled = anv_gem_get_bit6_swizzle(fd, I915_TILING_X);
-
/* Starting with Gen10, the timestamp frequency of the command streamer may
* vary from one part to another. We can query the value from the kernel.
*/
device->info.gen < 8 || !device->has_context_isolation;
device->compiler->supports_shader_constants = true;
+ /* Broadwell PRM says:
+ *
+ * "Before Gen8, there was a historical configuration control field to
+ * swizzle address bit[6] for in X/Y tiling modes. This was set in three
+ * different places: TILECTL[1:0], ARB_MODE[5:4], and
+ * DISP_ARB_CTL[14:13].
+ *
+ * For Gen8 and subsequent generations, the swizzle fields are all
+ * reserved, and the CPU's memory controller performs all address
+ * swizzling modifications."
+ */
+ bool swizzled =
+ device->info.gen < 8 && anv_gem_get_bit6_swizzle(fd, I915_TILING_X);
+
isl_device_init(&device->isl_dev, &device->info, swizzled);
result = anv_physical_device_init_uuids(device);
VkPhysicalDevice physicalDevice,
VkPhysicalDeviceFeatures2* pFeatures)
{
+ ANV_FROM_HANDLE(anv_physical_device, pdevice, physicalDevice);
anv_GetPhysicalDeviceFeatures(physicalDevice, &pFeatures->features);
vk_foreach_struct(ext, pFeatures->pNext) {
case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_16BIT_STORAGE_FEATURES: {
VkPhysicalDevice16BitStorageFeatures *features =
(VkPhysicalDevice16BitStorageFeatures *)ext;
- ANV_FROM_HANDLE(anv_physical_device, pdevice, physicalDevice);
-
features->storageBuffer16BitAccess = pdevice->info.gen >= 8;
features->uniformAndStorageBuffer16BitAccess = pdevice->info.gen >= 8;
features->storagePushConstant16 = pdevice->info.gen >= 8;
break;
}
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_BUFFER_ADDRESS_FEATURES_EXT: {
+ VkPhysicalDeviceBufferAddressFeaturesEXT *features = (void *)ext;
+ features->bufferDeviceAddress = pdevice->use_softpin &&
+ pdevice->info.gen >= 8;
+ features->bufferDeviceAddressCaptureReplay = false;
+ features->bufferDeviceAddressMultiDevice = false;
+ break;
+ }
+
case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_CONDITIONAL_RENDERING_FEATURES_EXT: {
VkPhysicalDeviceConditionalRenderingFeaturesEXT *features =
(VkPhysicalDeviceConditionalRenderingFeaturesEXT*)ext;
break;
}
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_DEPTH_CLIP_ENABLE_FEATURES_EXT: {
+ VkPhysicalDeviceDepthClipEnableFeaturesEXT *features =
+ (VkPhysicalDeviceDepthClipEnableFeaturesEXT *)ext;
+ features->depthClipEnable = VK_TRUE;
+ break;
+ }
+
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_INLINE_UNIFORM_BLOCK_FEATURES_EXT: {
+ VkPhysicalDeviceInlineUniformBlockFeaturesEXT *features =
+ (VkPhysicalDeviceInlineUniformBlockFeaturesEXT *)ext;
+ features->inlineUniformBlock = true;
+ features->descriptorBindingInlineUniformBlockUpdateAfterBind = false;
+ break;
+ }
+
case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_MULTIVIEW_FEATURES: {
VkPhysicalDeviceMultiviewFeatures *features =
(VkPhysicalDeviceMultiviewFeatures *)ext;
break;
}
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_YCBCR_IMAGE_ARRAYS_FEATURES_EXT: {
+ VkPhysicalDeviceYcbcrImageArraysFeaturesEXT *features =
+ (VkPhysicalDeviceYcbcrImageArraysFeaturesEXT *)ext;
+ features->ycbcrImageArrays = VK_TRUE;
+ break;
+ }
+
default:
anv_debug_ignored_stype(ext->sType);
break;
16 * devinfo->max_cs_threads,
16 * devinfo->max_cs_threads,
},
- .subPixelPrecisionBits = 4 /* FIXME */,
+ .subPixelPrecisionBits = 8,
.subTexelPrecisionBits = 4 /* FIXME */,
.mipmapPrecisionBits = 4 /* FIXME */,
.maxDrawIndexedIndexValue = UINT32_MAX,
break;
}
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_EXTERNAL_MEMORY_HOST_PROPERTIES_EXT: {
+ VkPhysicalDeviceExternalMemoryHostPropertiesEXT *props =
+ (VkPhysicalDeviceExternalMemoryHostPropertiesEXT *) ext;
+ /* Userptr needs page aligned memory. */
+ props->minImportedHostPointerAlignment = 4096;
+ break;
+ }
+
case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_ID_PROPERTIES: {
VkPhysicalDeviceIDProperties *id_props =
(VkPhysicalDeviceIDProperties *)ext;
break;
}
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_INLINE_UNIFORM_BLOCK_PROPERTIES_EXT: {
+ VkPhysicalDeviceInlineUniformBlockPropertiesEXT *props =
+ (VkPhysicalDeviceInlineUniformBlockPropertiesEXT *)ext;
+ props->maxInlineUniformBlockSize = MAX_INLINE_UNIFORM_BLOCK_SIZE;
+ props->maxPerStageDescriptorInlineUniformBlocks =
+ MAX_INLINE_UNIFORM_BLOCK_DESCRIPTORS;
+ props->maxPerStageDescriptorUpdateAfterBindInlineUniformBlocks =
+ MAX_INLINE_UNIFORM_BLOCK_DESCRIPTORS;
+ props->maxDescriptorSetInlineUniformBlocks =
+ MAX_INLINE_UNIFORM_BLOCK_DESCRIPTORS;
+ props->maxDescriptorSetUpdateAfterBindInlineUniformBlocks =
+ MAX_INLINE_UNIFORM_BLOCK_DESCRIPTORS;
+ break;
+ }
+
case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_MAINTENANCE_3_PROPERTIES: {
VkPhysicalDeviceMaintenance3Properties *props =
(VkPhysicalDeviceMaintenance3Properties *)ext;
props->maxTransformFeedbackStreamDataSize = 128 * 4;
props->maxTransformFeedbackBufferDataSize = 128 * 4;
props->maxTransformFeedbackBufferDataStride = 2048;
- props->transformFeedbackQueries = VK_FALSE;
+ props->transformFeedbackQueries = VK_TRUE;
props->transformFeedbackStreamsLinesTriangles = VK_FALSE;
props->transformFeedbackRasterizationStreamSelect = VK_FALSE;
props->transformFeedbackDraw = VK_TRUE;
anv_gem_munmap(map, device->hiz_clear_bo.size);
}
+static bool
+get_bo_from_pool(struct gen_batch_decode_bo *ret,
+ struct anv_block_pool *pool,
+ uint64_t address)
+{
+ for (uint32_t i = 0; i < pool->nbos; i++) {
+ uint64_t bo_address = pool->bos[i].offset & (~0ull >> 16);
+ uint32_t bo_size = pool->bos[i].size;
+ if (address >= bo_address && address < (bo_address + bo_size)) {
+ *ret = (struct gen_batch_decode_bo) {
+ .addr = bo_address,
+ .size = bo_size,
+ .map = pool->bos[i].map,
+ };
+ return true;
+ }
+ }
+ return false;
+}
+
+/* Finding a buffer for batch decoding */
+static struct gen_batch_decode_bo
+decode_get_bo(void *v_batch, bool ppgtt, uint64_t address)
+{
+ struct anv_device *device = v_batch;
+ struct gen_batch_decode_bo ret_bo = {};
+
+ assert(ppgtt);
+
+ if (get_bo_from_pool(&ret_bo, &device->dynamic_state_pool.block_pool, address))
+ return ret_bo;
+ if (get_bo_from_pool(&ret_bo, &device->instruction_state_pool.block_pool, address))
+ return ret_bo;
+ if (get_bo_from_pool(&ret_bo, &device->binding_table_pool.block_pool, address))
+ return ret_bo;
+ if (get_bo_from_pool(&ret_bo, &device->surface_state_pool.block_pool, address))
+ return ret_bo;
+
+ if (!device->cmd_buffer_being_decoded)
+ return (struct gen_batch_decode_bo) { };
+
+ struct anv_batch_bo **bo;
+
+ u_vector_foreach(bo, &device->cmd_buffer_being_decoded->seen_bbos) {
+ /* The decoder zeroes out the top 16 bits, so we need to as well */
+ uint64_t bo_address = (*bo)->bo.offset & (~0ull >> 16);
+
+ if (address >= bo_address && address < bo_address + (*bo)->bo.size) {
+ return (struct gen_batch_decode_bo) {
+ .addr = bo_address,
+ .size = (*bo)->bo.size,
+ .map = (*bo)->bo.map,
+ };
+ }
+ }
+
+ return (struct gen_batch_decode_bo) { };
+}
+
VkResult anv_CreateDevice(
VkPhysicalDevice physicalDevice,
const VkDeviceCreateInfo* pCreateInfo,
if (!device)
return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
+ const unsigned decode_flags =
+ GEN_BATCH_DECODE_FULL |
+ ((INTEL_DEBUG & DEBUG_COLOR) ? GEN_BATCH_DECODE_IN_COLOR : 0) |
+ GEN_BATCH_DECODE_OFFSETS |
+ GEN_BATCH_DECODE_FLOATS;
+
+ gen_batch_decode_ctx_init(&device->decoder_ctx,
+ &physical_device->info,
+ stderr, decode_flags, NULL,
+ decode_get_bo, NULL, device);
+
device->_loader_data.loaderMagic = ICD_LOADER_MAGIC;
device->instance = physical_device->instance;
device->chipset_id = physical_device->chipset_id;
if (device->info.gen >= 10)
anv_device_init_hiz_clear_value_bo(device);
+ if (physical_device->use_softpin)
+ device->pinned_buffers = _mesa_pointer_set_create(NULL);
+
anv_scratch_pool_init(device, &device->scratch_pool);
anv_queue_init(device, &device->queue);
anv_queue_finish(&device->queue);
+ if (physical_device->use_softpin)
+ _mesa_set_destroy(device->pinned_buffers, NULL);
+
#ifdef HAVE_VALGRIND
/* We only need to free these to prevent valgrind errors. The backing
* BO will go away in a couple of lines so we don't actually leak.
anv_gem_destroy_context(device, device->context_id);
+ gen_batch_decode_ctx_finish(&device->decoder_ctx);
+
close(device->fd);
vk_free(&device->alloc, device);
mem->map = NULL;
mem->map_size = 0;
mem->ahw = NULL;
+ mem->host_ptr = NULL;
uint64_t bo_flags = 0;
goto success;
}
+ const VkImportMemoryHostPointerInfoEXT *host_ptr_info =
+ vk_find_struct_const(pAllocateInfo->pNext,
+ IMPORT_MEMORY_HOST_POINTER_INFO_EXT);
+ if (host_ptr_info && host_ptr_info->handleType) {
+ if (host_ptr_info->handleType ==
+ VK_EXTERNAL_MEMORY_HANDLE_TYPE_HOST_MAPPED_FOREIGN_MEMORY_BIT_EXT) {
+ result = vk_error(VK_ERROR_INVALID_EXTERNAL_HANDLE);
+ goto fail;
+ }
+
+ assert(host_ptr_info->handleType ==
+ VK_EXTERNAL_MEMORY_HANDLE_TYPE_HOST_ALLOCATION_BIT_EXT);
+
+ result = anv_bo_cache_import_host_ptr(
+ device, &device->bo_cache, host_ptr_info->pHostPointer,
+ pAllocateInfo->allocationSize, bo_flags, &mem->bo);
+
+ if (result != VK_SUCCESS)
+ goto fail;
+
+ mem->host_ptr = host_ptr_info->pHostPointer;
+ goto success;
+ }
+
/* Regular allocate (not importing memory). */
if (export_info && export_info->handleTypes)
}
}
+VkResult anv_GetMemoryHostPointerPropertiesEXT(
+ VkDevice _device,
+ VkExternalMemoryHandleTypeFlagBits handleType,
+ const void* pHostPointer,
+ VkMemoryHostPointerPropertiesEXT* pMemoryHostPointerProperties)
+{
+ ANV_FROM_HANDLE(anv_device, device, _device);
+
+ assert(pMemoryHostPointerProperties->sType ==
+ VK_STRUCTURE_TYPE_MEMORY_HOST_POINTER_PROPERTIES_EXT);
+
+ switch (handleType) {
+ case VK_EXTERNAL_MEMORY_HANDLE_TYPE_HOST_ALLOCATION_BIT_EXT: {
+ struct anv_physical_device *pdevice = &device->instance->physicalDevice;
+
+ /* Host memory can be imported as any memory type. */
+ pMemoryHostPointerProperties->memoryTypeBits =
+ (1ull << pdevice->memory.type_count) - 1;
+
+ return VK_SUCCESS;
+ }
+ default:
+ return VK_ERROR_INVALID_EXTERNAL_HANDLE;
+ }
+}
+
void anv_FreeMemory(
VkDevice _device,
VkDeviceMemory _mem,
return VK_SUCCESS;
}
+ if (mem->host_ptr) {
+ *ppData = mem->host_ptr + offset;
+ return VK_SUCCESS;
+ }
+
if (size == VK_WHOLE_SIZE)
size = mem->bo->size - offset;
{
ANV_FROM_HANDLE(anv_device_memory, mem, _memory);
- if (mem == NULL)
+ if (mem == NULL || mem->host_ptr)
return;
anv_gem_munmap(mem->map, mem->map_size);
buffer->usage = pCreateInfo->usage;
buffer->address = ANV_NULL_ADDRESS;
+ if (buffer->usage & VK_BUFFER_USAGE_SHADER_DEVICE_ADDRESS_BIT_EXT) {
+ pthread_mutex_lock(&device->mutex);
+ _mesa_set_add(device->pinned_buffers, buffer);
+ pthread_mutex_unlock(&device->mutex);
+ }
+
*pBuffer = anv_buffer_to_handle(buffer);
return VK_SUCCESS;
if (!buffer)
return;
+ if (buffer->usage & VK_BUFFER_USAGE_SHADER_DEVICE_ADDRESS_BIT_EXT) {
+ pthread_mutex_lock(&device->mutex);
+ _mesa_set_remove_key(device->pinned_buffers, buffer);
+ pthread_mutex_unlock(&device->mutex);
+ }
+
vk_free2(&device->alloc, pAllocator, buffer);
}
+VkDeviceAddress anv_GetBufferDeviceAddressEXT(
+ VkDevice device,
+ const VkBufferDeviceAddressInfoEXT* pInfo)
+{
+ ANV_FROM_HANDLE(anv_buffer, buffer, pInfo->buffer);
+
+ assert(buffer->address.bo->flags & EXEC_OBJECT_PINNED);
+
+ return anv_address_physical(buffer->address);
+}
+
void
anv_fill_buffer_surface_state(struct anv_device *device, struct anv_state state,
enum isl_format format,